JPS6376018A - Controlling method for reason - Google Patents

Abstract

PURPOSE:To attain bidirectional rapid search by forming a link as a main body and forming network expression connecting a bidirectional pointer between a link corresponding to an intermediate result and a final conclusion and another link. CONSTITUTION:When a problem setting system body 1 detects the contradiction of a conclusion, the body 1 inquires the reason of an inference to a reason reference part 23 to be a submodule of a network control part 2 and the reference part 23 returns only leaves of the tree of the reason regarding a specified ink as its root by referring network expression 4 to the body 1 as a result. If the body 1 detects an error from the returned hypothesis or input data or when a user detects an error in the input data on the way of use and corrects the error, the body 1 commands an influence control part 25 to annual the inference. Then, the influence control part 25 deletes all the links up to the root along the tree of the reason regarding the specified links as leaves by referring the network expression 4 from a storage area. Thus, the tree of the reason can be rapidly and bidrectionally followed.

Description

[Detailed Description of the Invention] [Summary] A method for managing evidence, which uses a network consisting of nodes and links to obtain input data and intermediate conclusions.

In a system that solves a problem by expressing a hypothesis, final conclusion, etc. and making inferences, an inference is made based on the hypothesis or input data that is the basis for the conclusion, or conversely, based on the hypothesis or input data. In order to quickly refer to the conclusion, the network is represented mainly by links, and a two-way pointer is provided between the link corresponding to the intermediate or final conclusion and the link corresponding to its direct basis. Allows you to follow the chain quickly.

[Industrial Field of Application] The present invention relates to a method for managing evidence, and more specifically to a problem-solving system that uses a network of nodes and links to express input data, intermediate conclusions, hypotheses, final conclusions, etc. In particular, it relates to how evidence is managed within the network.

In this scale system, when the reasoning reaches a dead end, the previous hypothesis is overturned and the JIt theory is based on another hypothesis,
In addition, the input data, intermediate conclusions, and hypotheses that are the basis for the final conclusion are displayed according to the user's request, and if the user corrects the input data, the input data before the correction is used as the basis.
It is necessary to discard the conclusion that follows.

Therefore, there is a need for a method of quickly referencing a hypothesis or input data that is the basis for a conclusion, or conversely, a conclusion inferred from a hypothesis or input data based on that hypothesis or input data.

[Prior Art] FIG. 7 is a diagram showing an example of a network configuration, in which N1 to N
5 is a node, and Ll to L4 are links. N+ indicates a switch, N2 indicates a light, and nodes N and N2 are linked respectively.1. It is connected to the node N3 which indicates the tongue determination "No" via the node N3. Link Lt indicates a failure state, and link L3 indicates a disconnection state. Further, the node N1 is connected to the node N4 indicating the "ON" state via the link L2 indicating the setting position, and the node N2 is connected to the node N5 indicating the positive "YES" via the link L4 indicating the lighting state. It is connected.

Conventional problem-solving systems have used node-based expressions that allocate storage areas to each node in the network. FIG. 8 is a block diagram of a conventional problem-solving system, and FIG. 9 is a diagram showing an example of a node-based network representation for the network shown in FIG. 7.

FIG. 9 <a> shows the structure of the storage area allocated to the "switch" node, and the first column indicates that it is allocated to this node thickness "switch". In addition, from 2B onwards, "the current setting position of the switch is ON, and that information was given as input data (INP (JT)) from the user", and "the switch is not currently malfunctioning and The information each represents the information given as input data (INPUT) from the user.The fact that the 21st fl11 and 3rd tlifl are related to "setting position" and "failure" respectively is not managed separately in some way. has been done.

In FIG. 8, the problem solving system body 1 is configured such that, for example, [
When the reference unit 21, which is a sub-module of the network management unit 2, is inquired as to whether the switch is currently set to ON, the reference unit 21 refers to the network representation (based on nodes) 3, and as shown in FIG. Search for (a) and ``
Returns "true" as the result. "The switch is currently set to O.
"Is it FF?", returns "false". Since "the current switch setting position is ON", the problem-solving system main body 1 determines by inference that the light should be set to a point, and sends the update section 22, which is a submodule of the network management section 2, " Based on the fact that the switch is currently in the ON position, it is notified that it has been inferred that the light will turn on.

On the other hand, the updating unit 22 updates the "switch" node and the "light" node, respectively, as shown in FIG. 9(b).

Update as shown in (C). In Figure 9(b), the second column indicates that ``the light will turn on based on this information''.
"It was discussed" is added.

FIG. 9(C) shows the structure of the storage area allocated to the ``Light'' connode, and the 11'I indicates that this node is allocated to ``Light''. . Also, from the second column onwards, ``The light is currently on, and the reason for this is that ``the current setting position of the switch is ON,'' and ``The light is not currently disconnected, and that information is Each indicates that the data was given as input data from the user. The fact that the second and third columns are related to "light on" and "disconnection" respectively means that
managed separately in some way.

If the above-mentioned inference is executed in many stages, and if the problem-solving system 1 forms a hypothesis during the process, a field with an empty "Reason" field is formed in the network, and such a hypothesis or input A “evidence tree” with data as leaves and conclusions as roots
is formed. When the problem solving system body 1 discovers a contradiction in the conclusion.

It is necessary to trace the tree of evidence from the root to the leaves and discover incorrect hypotheses or questionable input data.

In that case, the problem-solving system main body 1 informs the root)\1 reference unit 23, which is a submodule of the network management unit 2, of the reason why it has come to infer that “the light is on”, for example. "What is it?" and the basis reference section 23
refers to network expression 3, searches for FIG. 9(C), and returns "the current switch setting position is ON" as a result. The problem solving system main body 1 repeats this operation until it reaches the leaf of the basis.

In this way, when a wrong hypothesis is discovered, or when the user notices a gap in the input data midway through and corrects it, the problem-solving system main body 1 makes inferences based on the wrong hypothesis and input data. All conclusions must be discarded. In that case, the problem solving system main body 1 includes a ripple reference section 24 which is a submodule of the network management section 2.
For example, ask them, ``What can be inferred from the fact that the switch is currently in the ON position?''
The ripple reference unit 24 refers to the network representation 3, searches for FIG. 9(b), and returns "Light is a point" as a result. The problem solving system main body 1 requests the updating unit 22 to delete this information and repeats this operation until it reaches the root of the basis tree.

[Problems to be solved by the invention] According to the conventional method, although the links in the network correspond to the information that should originally be expressed, such as input data, intermediate conclusions of inference, hypotheses, and final conclusions, nodes Because a network representation based on R is used, it is not possible to directly represent the ``evidence'', which is the relationship between these input data, the intermediate conclusion of the inference, the hypothesis, and the final conclusion of R, as a pointer in the storage area. , there was a flaw in that it was difficult to trace the basis tree at high speed.

The present invention has been made in view of these points, and
The purpose of this invention is to provide a basis management method that can quickly trace a basis tree in forward and backward directions.

[Means for Solving the Problems] FIG. 1 is a flowchart showing the principle of the present invention. The present invention first uses a network consisting of nodes and links to generate input data, intermediate conclusions, and hypotheses.

In a problem-solving system that expresses final conclusions and makes inferences, links are the main body, and bidirectional pointers are provided between links corresponding to intermediate and final conclusions and other links on which they are based. Next, based on the network representation, input data, intermediate conclusions, hypotheses, final conclusions, etc. are expressed and inference is made (step ①).

[Operation] By creating a network representation mainly based on links, rather than a network mainly based on nodes as in the conventional method, it is possible to give a storage area to each link, so the basis of the relationship between links can be directly used as a pointer. express. This allows the base tree to be directly expressed as a pointer in the storage area, making it possible to perform bidirectional high-speed searches from leaves to roots, or conversely from roots to leaves.

[Example] Hereinafter, an example of the present invention will be described in detail with reference to the drawings. As in the case of the conventional example, let us take as an example the case where the network shown in FIG. 7 is expressed. FIG. 2 shows the configuration diagram of the problem solving system according to the present invention, and FIG. FIG. 3 is a diagram showing an example of a network representation. Second
In the figure, the same parts as in FIG. 8 are denoted by the same reference numerals. FIG. The column is the name of the corresponding link, the second column is the source of the link, the third column is the destination of the link, the fourth column is the base address of the storage area that represents the other link that is the base of this link, and the fifth column is the base address of the storage area that represents the other link that is the base of this link. The columns represent destination addresses of storage areas representing other links inferred based on this link. This base address and destination address are described as ``bidirectional pointers,'' which enable bidirectional searches from the root to the vegetable or from the leaf to the root.

For the link corresponding to the input data, "IN" is displayed in the fourth column.
A value "PUT" is set, and a value "[HYPOJ" indicating the hypothesis is set in the fourth column for the link corresponding to the hypothesis. Therefore, in the example of Fig. 3, the current setting position of the switch is ON, and this information is given as input data (INPIJT) from the user, and based on this information, the light is turned on. ``It is inferred that the switch is currently not faulty and that information was provided as input data from the user (TNPtJT)'', ``The light is currently on and the reason is `` The current setting position of the switch is "ON", "
This indicates that the light is not currently disconnected and the information was given as input data (INPUT) from the user.

In this method, the problem solving system main body 1, for example, [
When the reference unit 21, which is a submodule of the network management unit 2, is inquired as to whether the switch is currently set to ON, the reference unit 21 refers to the network representation (mainly links) 4, and as shown in FIG. Search for and return address 1 as the result.

In response to "Is the current switch setting position OFF?", an appropriate value representing "false" (for example, 0) is returned. [Since the current switch setting position is ON, the problem-solving system body 1 uses inference to determine that the light will turn on, and tells the update unit 22, which is a submodule of the network management unit 2, to use address 1 as a basis. As a result, we will notify you that we have inferred that the light will do this at the point. In response, the updating unit 22 updates addresses 1 and 3 as shown in FIG. Problem solving system main body 1 during inference
If a hypothesis is made, the basis address field is r I-! Y
A link is generated that is P OJ.

The "ground reference unit 23" and the "ripple processing unit 25" perform unique operations in the system of the present invention compared to the conventional system. FIGS. 4 and 5 show respective operating algorithms.

If the problem-solving system body 1 discovers a contradiction in the conclusion,
The problem-solving system main body 1 sends information such as "
The basis for inferring address 3 is asked by J, and the basis reference unit 23 refers to network representation 4 to find the leaf of the basis tree whose root is the specified link, in this example, address 1 (the third (see figure) is returned as the result.

When the problem-solving system main body 1 discovers an error among the hypotheses and input data returned in this way, or when the user notices a mistake in the input data midway through and corrects it, the problem-solving system main body 1 is network management section 2
For example, the ripple management unit 25 instructs the ripple management unit 25, which is a submodule of
All links that follow the basis tree with the specified link as a leaf, up to the root, in this example, address 3.
Delete only the link from the storage area.

FIG. 6 is a concrete configuration diagram of a problem solving system according to the present invention. The entire system is roughly divided into a problem solving system body 1, a network management section 2, and a knowledge base 5. Here, the network representation 4 mainly consisting of links is considered to be included in the network management section 2.

The problem-solving system main body 1 is in charge of the interface with the user and problem-solving work by inference, and has an input section 11. Reasoning part 12. Hypothesis setting section 13. Contradiction resolution part 14. Correction iE processing unit 15. It consists of a knowledge reference section 1G and an output section 17.

The network management unit 2 inputs 'the input data, tit
Intermediate results, hypotheses, final conclusions, etc. of the theory are held as a network, and the reference section 21. Update section 22. Basis reference section 23. It consists of a ripple processing section 25 and a network representation 4. The knowledge base 5 is a database that stores knowledge used for inference.

The i-like operations of the problem-solving system main body 1 are as follows.

The input unit 11 () asks questions to the user, and (2) receives input data, corrections to past inputs, and requests for explanation from the user. ■The input data is transferred to the inference section 12, and when the past input is corrected, ■The correction processing section 15 is notified of this, and when an explanation of the inference process is requested, ■The basis reference section 23 says, The receiver takes the letter and presents it to the publisher.

The inference unit 12 performs inference from (1) input data and (2) already generated intermediate results and hypotheses, (2) uses knowledge provided by the knowledge reference unit 16, and (4) updates the network representation 4 according to the inference results. If further input data is required during inference, (1) requests the input unit 11 to ask a question; if the necessary data cannot be obtained, requests the 0-hypothesis setting unit 13 to formulate an appropriate hypothesis; Furthermore, if a contradiction is found, O contradiction resolution section 1
4 to that effect. Once the final conclusion is reached in this way, it is output from the O output section 17.

The hypothesis setting unit 13 generates an appropriate hypothesis and adds it to the 0 network representation 4. The contradiction resolution unit 14: ■ Finds the basis for the conclusion of the contradiction, and if it is input data, ■ Requests the usage name to make a correction, and if it is a hypothesis, requests the [phase] correction processing unit 15 to discard the hypothesis. do.

When the correction loco processing unit 15 receives a notification of data correction from the input unit 11 and a request to discard a hypothesis from the contradiction resolution unit 14, it deletes all input data and conclusions drawn from the hypothesis. ■ Notify the network management section 2 of this fact.

The knowledge reference unit 16 (1) extracts knowledge necessary for inference from the knowledge base 5 and (2) transfers it to the inference unit 12. The output unit 17 presents the final conclusion to the usage fee.

[Effects of the Invention] As explained in detail above, according to the present invention, a network consisting of nodes and links can be changed from a network representation mainly composed of nodes to a network representation mainly composed of links that themselves have information. In addition, by adding bidirectional pointers between links corresponding to intermediate and final conclusions and other links on which they are based, the evidence tree can be moved forward and backward at high speed. The practical effects are extremely large.

[Brief explanation of the drawing]

Fig. 1 is a flowchart showing the principle of the present invention, Fig. 2 is a configuration diagram of a problem solving system according to the invention, Fig. 3 is a diagram showing an example of network representation mainly based on links, Fig. 4 is a reference to the rationale. FIG. 5 is a diagram showing the algorithm of the ripple processing section, FIG. 6 is a specific configuration diagram of the problem solving system according to the present invention, FIG. 7 is a diagram illustrating an example of the network configuration, and FIG. This figure is a block diagram of a conventional problem-solving system, and FIG. 9 is a diagram showing an example of a conventional network representation mainly consisting of nodes. In Figures 2 and 6, 1 is the problem solving system itself, 2 is the network management section, 4 is the network representation, and 5 is the knowledge base. ! Figure 1 shows the first principle of the invention. Crude answer for the present invention.
Example 2 of 5-column diagonal body and ζ component work representation TI
ZI Figure 3 shows the alf'rhythm of the material reference section, and the alcopy of the inscription processing section with four sections l shows the λ' mode.Figure 5 shows the usage book i! Theory of the present invention Clear method system l subhia area (7th 1) (b) r switch j note attached to h Figure 9.

Claims (1)

[Claims] A problem-solving system that expresses input data, intermediate conclusions, hypotheses, and final conclusions and makes inferences using a network made up of nodes and links, which is based on links and corresponds to intermediate and final conclusions. A network representation is created in which a bidirectional pointer is provided between the link to be used and other links on which it is based (step [1]), and based on the network representation, input data, intermediate conclusions, hypotheses, Express the final conclusion and make inferences (step [2
]), a basis management method characterized by the following.